SPEECH RECOGNITION

The Speech recognition technology, also known as Voice recognition, is used to turn into a text the acoustic signal corresponding to what a person is saying. In recent years this technology is recovering interest from the developers and users due to the following reasons, among others: in the social field, the conversion of a voice into a text is a key to increase the accessibility of the computer systems; from the managerial point of view, the demand for subtitled audiovisual contents is increasing constantly, so that the manual transcriptions are inefficient and expensive; and in the Internet field, it is more and more necessary to have audiovisual contents processing that go beyond the text.

Any Voice Recognition system is based on two main elements: the acoustic model, that makes it possible to make the acoustic signal of the input close to the ones corresponding to the language phonemes, and the language model, that corrects possible mistakes made by the acoustic model, rejecting structures that are not valid in the language. Besides we distinguish focuses directed to the speaker that require a specific training of the models for the people's voices you want to recognize and the speakers' independent focuses, valid for any speaker without needing to previously train.

At Daedalus we are especially interested in the speaker's independent processes and, on one hand, we apply our technology and linguistic knowledge to improve the language models of the Voice Recognition systems; on the other hand, we integrate this type of technology in Information retrieval systems, making possible the search on audio and video contents. Besides, we use this technology in the process of automatic generation of subtitles for audiovisual contents, which reduces time and costs. The STILUS Subtitler product incorporates the Daedalus technology of generation of subtitles.

Daedalus works with third-party voice recognition technologies: commercial products, such as the Sail Labs or Nuance ones, as well as, mostly in the field of research projects, open source products, such as HTK or Sphinx. At Daedalus we adapt these systems to our clients' business environments and fields.

The Digital Audio Library Indexing (DALI) demo is included in our Showroom.

Do you want more information on STILUS Subtitler, the Daedalus product for automatic generation of subtitles?